attribute describes observations #583 revisit

[canwaf]

Instead of requiring that all attributes need to be attached to an observation column in a pivoted dataset, attributes which are not so attached should be present on all observations.

The current mechanism for attaching attributtions to observations is in observations.py.

For attributes in a pivoted dataset we should support the following.

• "describes_observations": "a value" single observation column, applies to only that column
• "describes_observations": False the column applies to all observations

Update the error message for when describes_observations is either not set in pivoted datasets, or doesn't match a column, the user is told to make sure the column is defined correctly, or the the value to false so that the attribute is applicable to all observations.